Short version: Strip away the AEO/GEO mystique, and what OpenAI and Google actually say officially comes down to fairly ordinary fundamentals — stay crawlable, answer questions directly, build real authority — not secret schema types or special AI files. I've found it genuinely clarifying to go check primary sources instead of another blog's checklist, and to see how this actually plays out for a real brand, Obsurfable's Explorer is a public corpus of real recorded AI answers worth searching before assuming anything based on theory.
What OpenAI actually says, officially
OpenAI's own publisher documentation is more specific than most secondhand advice makes it sound. Any public site can appear in ChatGPT's search results, and the main technical requirement is not blocking OAI-SearchBot — the specific crawler responsible for search inclusion and citations, separate from GPTBot, which handles training data, and ChatGPT-User, which fetches a page only when someone directly asks ChatGPT to read it. These three can be allowed or blocked independently in robots.txt, and treating them as one undifferentiated "AI bot" is a common, avoidable mistake. OpenAI also confirms that sites allowing OAI-SearchBot can track referral traffic through standard analytics, since ChatGPT search referrals carry a recognizable UTM parameter.
Worth noting: OpenAI is explicit that placement in search results isn't guaranteed even when a site is fully crawlable — inclusion depends on a mix of ranking factors, not a checklist that unlocks appearance automatically. OpenAI also notes that accessibility markup — the same ARIA labels and roles that support screen readers — helps its ChatGPT Atlas agent understand page structure and interactive elements more reliably, which is a detail that rarely shows up in generic AEO advice at all.
What Google actually says, officially
Google published updated generative-AI-features guidance in May 2026, and it's more blunt than a lot of AEO advice would suggest. According to Google's own documentation, sites don't need to create llms.txt files, AI-specific text formats, or Markdown versions of pages to appear in AI Overviews or AI Mode — Google's systems may crawl those files, but they aren't given any special treatment. The same guidance says there's no dedicated schema.org markup required for generative AI visibility, no need to break content into small "chunks" for AI systems to understand it, and no need to rewrite content specifically for AI phrasing, since Google's systems already handle synonyms and general meaning. Google frames AEO and GEO as functionally still SEO for its own search surfaces, running on the same index and ranking systems as standard results.
Google's guidance also directly warns against a specific tactic that's easy to fall into once you know AI systems answer narrow questions: building a separate page for every conceivable sub-query variation of the same topic. Doing that primarily to influence rankings or generative answers falls under Google's scaled-content-abuse policy, which is a useful boundary to know before treating "cover every possible phrasing" as an unqualified best practice.
Where the advice and the official guidance actually diverge
| Common AEO/GEO tactic | What OpenAI says | What Google says |
|---|---|---|
| llms.txt files | Not mentioned as a requirement | Explicitly not needed; ignored for ranking purposes |
| Special AI-specific schema | Not addressed directly | Explicitly says no special schema.org type exists for this |
| Allowing/blocking specific crawlers | Explicitly matters — OAI-SearchBot access is the key control | Standard crawlability rules apply, no AI-specific crawler distinction described |
| Breaking content into small chunks | Not addressed | Explicitly says not required |
| Rewriting content in "AI-friendly" phrasing | Not addressed | Explicitly says not necessary |
Seeing it laid out this way was useful for me mainly because it cut down on second-guessing. A fair number of tactics circulating as essential AEO/GEO practice simply aren't confirmed as mattering by the two companies actually running the systems in question. That doesn't automatically make them useless in every context — a technical documentation site might still find value in an llms.txt file for reasons unrelated to Google or ChatGPT ranking, for instance — but it does mean treating them as required steps for citation isn't backed by what either company has actually published.
What still matters, according to both
Where OpenAI and Google actually agree: crawlability is foundational, content needs to genuinely and directly answer real questions, and technical basics — proper indexing, no accidental noindex, reasonable page speed — remain a prerequisite rather than an optional extra. Neither company is describing some separate, exotic discipline distinct from doing the fundamentals well. The gimmicks that get sold as AEO-specific tricks mostly aren't part of either company's own account of how their systems work.
The one place OpenAI is more specific than Google
If there's a genuinely AI-specific technical step worth taking seriously, it's the crawler-level control OpenAI describes — deliberately confirming OAI-SearchBot isn't disallowed, distinct from a blanket "allow all bots" approach, and understanding that a page can still show up as a bare link and title even when disallowed, if referenced elsewhere, unless a noindex tag is used instead. That's a more granular and AI-platform-specific instruction than anything in Google's current guidance, which mostly folds AI visibility into standard Search Console-monitored fundamentals.
None of this means there's nothing left to build
Confirming that llms.txt and special schema aren't required doesn't mean the underlying work is done — it just narrows down where that work should actually go. Writing directly, building real topical depth, and earning genuine third-party mentions are still the parts neither company is offering a shortcut around.
What's still genuinely missing from official guidance on either side is any way to check whether it's working for your specific brand — that's not something OpenAI's or Google's documentation is trying to solve. Obsurfable's free AI visibility checker covers that gap directly, running real buyer-style questions and showing what comes back for your company specifically. And since a lot of citation behavior traces back to which domains a model already trusts, this report on the domains most frequently cited across LLMs is worth a look before assuming your own site's authority stacks up the way you think it does.
What I'd actually stop worrying about, and what I'd keep doing
Stop treating llms.txt or a special AI schema as a required checklist item — neither company's own guidance backs that up, at least not yet. Keep doing the fundamentals: crawlability, direct answers, real authority, and confirming OAI-SearchBot specifically isn't blocked if ChatGPT visibility matters to you. And keep checking actual results rather than assuming compliance with a checklist equals success — Obsurfable's Explorer is free to search directly, and since platforms genuinely behave differently under the hood, this comparison of ChatGPT and Perplexity's separate approaches is worth reading before assuming one playbook covers every system evenly.
FAQ
Do I actually need an llms.txt file to be cited by ChatGPT or Google's AI features? Not according to either company's own current guidance. Google explicitly says it's not required and is ignored for ranking purposes; OpenAI's publisher guidance doesn't identify it as a factor either.
Is there special schema markup required for AI Overviews or ChatGPT citations? No. Google's own documentation explicitly states there's no dedicated schema.org type needed for generative AI visibility — standard structured data still helps with normal search features, but it isn't a special AI unlock.
What's the one AI-specific technical step that does matter? Making sure OAI-SearchBot specifically isn't disallowed in robots.txt, if ChatGPT visibility matters to you — it's a distinct crawler from the one used for model training, and blocking it removes you from ChatGPT's search results and citations.
Does ranking well in Google guarantee I'll be cited by ChatGPT too? No. The two systems run separate crawling and ranking infrastructure, and OpenAI's and Google's own guidance describe different specific requirements — doing well on one doesn't guarantee anything on the other.
If the checklist tactics aren't required, what should I actually spend time on? The fundamentals both companies do emphasize: crawlability, content that answers real questions directly, and genuine authority — then check your actual results rather than assuming compliance with a checklist worked.
Does this mean AEO/GEO as a discipline is basically a myth? Not entirely — the underlying goal, being understood and cited accurately by AI systems, is real and worth pursuing. What's overstated is the idea that it requires an entirely separate, exotic set of tactics distinct from doing solid fundamentals well and then confirming they're working.
A fair amount of AEO/GEO advice online is built on inference about how these systems probably work, not on what the companies running them have actually published. Going to the primary source cut through more noise for me than another checklist did — the honest version of "how do I get cited" turns out to be closer to solid fundamentals, checked directly, than a growing pile of AI-specific hacks. Both companies can update this guidance again, so it's worth treating July 2026 as a snapshot rather than a permanent answer.
Comments
Loading comments…